ResearchBenchmarksOffline Reinforcement Learning on OGBench cube-triple-play (5 tasks)Follow14Success RateValue Flows0.483.997.511.01May 7, 2026Evaluation ResultsMethodMethodLinksSuccess RateNormalized ReturnValue FlowsSeeds=8Seeds=82026.0514—IQNSeeds=8Seeds=82026.056—FQLSeeds=8Seeds=82026.054—PCBFSeeds=8Seeds=82026.054—CODACSeeds=8Seeds=82026.052—IQLSeeds=8Seeds=82026.051—